How You Actually Build a Qubit

A qubit needs to be a physical system with two distinguishable states, controllable from outside, and well isolated from everything else. Those requirements conflict, since controlling something means coupling to it, and coupling to it means coupling to noise.

Different technologies resolve that conflict differently.

Superconducting qubits

Tiny circuits of superconducting metal, cooled to around 10 to 15 millikelvin, in which current can flow in a superposition of directions. Pursued by Google, IBM and others, and currently the most prominent approach.

Advantages. Very fast gates, measured in tens of nanoseconds. Built with fabrication techniques adapted from the semiconductor industry, so scaling to many qubits is comparatively tractable.

Disadvantages. Short coherence times, typically tens to hundreds of microseconds. Requires a dilution refrigerator, which is large, expensive and power hungry. Connectivity is limited to neighbouring qubits on a chip.

Trapped ions

Individual charged atoms held in place by electromagnetic fields in a vacuum, manipulated with lasers. Pursued by Quantinuum, IonQ and others.

Advantages. Excellent coherence times, reaching seconds. Every ion is identical, since they are atoms, so there is no manufacturing variation. Typically full connectivity, meaning any qubit can interact directly with any other.

Disadvantages. Gates are far slower, thousands of times slower than superconducting. Scaling to very large numbers in a single trap is difficult, so architectures rely on shuttling ions or linking multiple traps.

Neutral atoms

Uncharged atoms held by focused laser beams in optical tweezers, arranged in configurable arrays. A newer approach that has advanced quickly, pursued by QuEra, Pasqal and others.

Advantages. Large numbers of atoms can be arranged, and the geometry is reconfigurable, which is useful for both simulation and error correcting codes. Identical atoms, as with ions.

Disadvantages. Gate fidelities have historically trailed the more established approaches, though the gap has narrowed considerably.

Photonic qubits

Individual particles of light. Pursued by PsiQuantum, Xanadu and others.

Advantages. Photons interact weakly with their environment, so coherence is excellent and much of the system can operate without extreme refrigeration.

Disadvantages. The same weak interaction makes two qubit gates hard, since photons do not naturally interact with each other. Generating single photons on demand and detecting them reliably are both difficult.

Others

Spin qubits in silicon use the spin of an electron in a semiconductor structure, attractive because they could exploit existing chip manufacturing at very small size, though they remain less mature.

Topological qubits aim to encode information in a way that is intrinsically protected from local noise, which would dramatically reduce error correction overhead. Pursued principally by Microsoft. Scientifically ambitious and the underlying physics has been contested, so it should be regarded as high risk with a high potential payoff.

Bosonic and cat qubits encode information in states of a resonator chosen so that one error type is naturally suppressed, reducing what error correction must handle.

Why connectivity matters more than it sounds

An underappreciated practical factor.

If a qubit can only interact with its immediate neighbours, then an algorithm needing two distant qubits to interact must first move information across the chip through a chain of intermediate operations. Each of those consumes time and introduces error.

The consequence is that two machines with identical qubit counts and identical gate fidelities can differ substantially in what they can actually run, because one spends much of its available depth shuffling information and the other does not.

This is why full connectivity is a genuine advantage for trapped ion systems, and it partly offsets their much slower gates. Comparisons based on qubit count alone miss it entirely.

Why nobody has won

Because the requirements conflict, and each approach trades differently.

The metric that matters is roughly how many high quality operations can be performed before decoherence, combined with how many qubits can be assembled and how well they are connected. Superconducting qubits are fast but noisy and poorly connected. Trapped ions are accurate and well connected but slow and hard to scale. Photonics avoids refrigeration but struggles with gates.

None of these is obviously fatal and none is obviously winning. It is also possible that different approaches will serve different purposes, with one technology suited to error corrected general computation and another to specialised simulation.

For anyone outside the field the practical implication is to be sceptical of claims that one approach has definitively won, and to notice that a company's assessment of which approach is best reliably matches the approach it has chosen.